1. Field of the Invention
The present invention relates to an image compression technique that compresses an input multivalued image.
2. Description of the Related Art
In recent years, electronic processing of a paper document has been advanced with the widespread scanners. In general, a color image has a large file size. Therefore, a method for compressing an image by Joint Photographic Experts Group (JPEG) compression is currently widespread. Although the JPEG compression is greatly effective to compress a natural image such as a photo, image deterioration referred to as “mosquito noise” occurs in a character portion with the JPEG compression. Hence, methods are proposed as described in Japanese Patent Application Laid-Open Nos. 2004-260327 and 2005-012768. With such methods, an input image is divided into three regions, which are a character region, a photo region, and a background region. Further, the character region is binarized, and is then compressed by a modified modified read (MMR) method, and the background region is compressed by the JPEG method, thereby expressing a full color image with a smaller file size while keeping the quality of the character region.
In order to further reduce the file size as compared with the compression methods discussed in Japanese Patent Application Laid-Open Nos. 2004-260327 and 2005-012768, there may be a method to significantly reduce the size of the background region. For example, if the input image has 300 dpi, the file size can be greatly reduced by reducing the background region to 50 dpi or 25 dpi. If contents of the background region are not so important, it is advantageous to greatly reduce a resolution of the background region (e.g., 25 dpi). However, as discussed in Japanese Patent Application Laid-Open No. 2004-260327, if the resolution of the input image is significantly reduced before extracting a representative color of a character, the representative color is extracted using a reduced multivalued image with 50 dpi or 25 dpi, and thus a problem is caused that the representative color of the character cannot be properly extracted. The representative color of the character is not properly extracted because colors of a matrix having 12×12 pixels in an input image having 300 dpi are combined to those of one pixel in a reduced multivalued image having 25 dpi, and the character color is mixed with another background color. That is, if the colors with 12×12 pixels contain a plurality of colors, the colors are combined to one, which leads to a substantial lack of color information.
As discussed in Japanese Patent Application Laid-Open No. 2005-012768, when the image is reduced in resolution just before the JPEG compression, the representative colors of characters can be properly obtained with a color multivalued image therebefore. However, in processing for extracting the representative colors, a large memory capacity is required to temporarily store the color multivalued image before reducing the image in resolution. In the case of a color multivalued image with 300 dpi and 24-bit red-green-blue (RGB), 24 M bytes are required as a memory capacity. Therefore, the processing is not realized in an image compression apparatus with a low memory capacity. To increase the memory capacity, a problem is caused that costs of hardware seriously rise.